Three-dimensional vertical NOR flash thing-film transistor strings

[0001] The present invention is a continuation application of U.S. patent application (“Parent Application I”), Ser. No. 15/837,734, entitled “Three-dimensional vertical NOR Flash Thin film Transistor Strings,” filed on Dec. 11, 2017, which is a divisional application of U.S. patent application (“Parent Application II”), Ser. No. 15/343,332, entitled “Three-dimensional vertical NOR Flash Thin film Transistor Strings,” filed on Nov. 4, 2016, which is related to and claims priority of (i) U.S. provisional patent application (“Provisional Application I”), Ser. No. 62/260,137, entitled “Three-dimensional Vertical NOR Flash Thin-film Transistor Strings,” filed on Nov. 25, 2015; (ii) U.S. non-provisional patent application (“Non-Provisional Application I”), Ser. No. 15/220,375, “Multi-Gate NOR Flash Thin-film Transistor Strings Arranged in Stacked Horizontal Active Strips With Vertical Control Gates,” filed on Jul. 26, 2016; and (iii) U.S. provisional patent application (“Copending Provisional Application II”), Ser. No. 62/363,189, entitled “Capacitive Coupled Non-Volatile Thin-film Transistor Strings,” filed Jul. 15, 2016; and (iv) U.S. non-provisional patent application (“Non-Provisional Patent Application II”), Ser. No. 15/248,420, entitled “Capacitive Coupled Non-Volatile Thin-film Transistor Strings in Three-Dimensional Array,” filed Aug. 26, 2016. The disclosures of Parent Application I, Parent Application II, Provisional Application I, Provisional Application II, Non-Provisional Patent Application I, and Non-Provisional Patent Application II are hereby incorporated by reference in their entireties.

[0002] 1. Field of the Invention

[0003] The present invention relates to high-density memory structures. In particular, the present invention relates to high-density memory structures formed by interconnected thin film storage elements, such as thin film storage transistors formed in vertical strips with horizontal word lines.

[0004] 2. Discussion of the Related Art

[0005] In this disclosure, memory circuit structures are described. These structures may be fabricated on planar semiconductor substrates (e.g., silicon wafers) using conventional fabrication processes. To facilitate clarity in this description, the term “vertical” refers to the direction perpendicular to the surface of a semiconductor substrate, and the term “horizontal” refers to any direction that is parallel to the surface of that semiconductor substrate.

[0006] A number of high-density non-volatile memory structures, such as “three-dimensional vertical NAND strings,” are known in the prior art. Many of these high-density memory structures are formed using thin film storage transistors formed out of deposited thin films (e.g., polysilicon thin films), and organized as arrays of “memory strings.” One type of memory strings is referred to as NAND memory strings or simply “NAND strings”. A NAND string consists of a number of series-connected thin film storage transistors (“TFTs”). Reading or programming the content of any of the series-connected TFTs requires activation of all series-connected TFTs in the string. Thin film NAND transistors have lower conductivity than NAND transistors formed in single crystal silicon, therefore the low read current that is required to be conducted through a long string of NAND TFTs results in a relatively slow read access (i.e. long latency).

[0007] Another type of high density memory structure is referred to as the NOR memory strings or “NOR strings.” A NOR string includes a number of storage transistors each of which is connected to a shared source region and a shared drain region. Thus, the transistors in a NOR string are connected in parallel, so that a read current in a NOR string is conducted over a much lesser resistance than the read current through a NAND string. To read or program a storage transistor in a NOR string, only that storage transistor needs to be activated (i.e., “on” or conducting), all other storage transistors in the NOR string may remain dormant (i.e., “off” or non-conducting). Consequently, a NOR string allows much faster sensing of the activated storage transistor to be read. Conventional NOR transistors are programmed by a channel hot-electron injection technique, in which electrons are accelerated in the channel region by a voltage difference between the source region and the drain region and are injected into the charge-trapping layer between the control gate and the channel region, when an appropriate voltage is applied to the control gate. Channel hot-electron injection programming requires a relatively large electron current to flow through the channel region, therefore limiting the number of transistors that can be programmed in parallel. Unlike transistors that are programmed by hot-electron injection, in transistors that are programmed by Fowler-Nordheim tunneling or by direct tunneling, electrons are injected from the channel region to the charge-trapping layer by a high electric field that is applied between the control gate and the source and drain regions. Fowler-Nordheim tunneling and direct tunneling are orders of magnitude more efficient than channel hot-electron injection, allowing massively parallel programming; however, such tunneling is more susceptible to program-disturb conditions.

[0008] 3-Dimensional NOR memory arrays are disclosed in U.S. Pat. No. 8,630,114 to H. T Lue, entitled “Memory Architecture of 3D NOR Array,” filed on Mar. 11, 2011 and issued on Jan. 14, 2014.

[0009] U.S. patent Application Publication US2016/0086970 A1 by Haibing Peng, entitled “Three-Dimensional Non-Volatile NOR-type Flash Memory,” filed on Sep. 21, 2015 and published on Mar. 24, 2016, discloses non-volatile NOR flash memory devices consisting of arrays of basic NOR memory groups in which individual memory cells are stacked along a horizontal direction parallel to the semiconductor substrate with source and drain electrodes shared by all field effect transistors located at one or two opposite sides of the conduction channel.

[0010] Three-dimensional vertical memory structures are disclosed, for example, in U.S. Pat. No. 8,878,278 to Alsmeier et al. (“Alsmeier”), entitled “Compact Three Dimensional Vertical NAND and Methods of Making Thereof,” filed on Jan. 30, 2013 and issued on Nov. 4, 2014. Alsmeier discloses various types of high-density NAND memory structures, such as “terabit cell array transistor” (TCAT) NAND arrays (FIG. 1A), “pipe-shaped bit-cost scalable” (P-BiCS) flash memory (FIG. 1B) and a “vertical NAND” memory string structure. Likewise, U.S. Pat. No. 7,005,350 to Walker et al. (“Walker I”), entitled “Method for Fabricating Programmable Memory Array Structures Incorporating Series—Connected Transistor Strings,” filed on Dec. 31, 2002 and issued on Feb. 28, 2006, also discloses a number of three-dimensional high-density NAND memory structures.

[0011] U.S. Pat. No. 7,612,411 to Walker (“Walker II”), entitled “Dual-Gate Device and Method” filed on Aug. 3, 2005 and issued on Nov. 3, 2009, discloses a “dual gate” memory structure, in which a common active region serves independently controlled storage elements in two NAND strings formed on opposite sides of the common active region.

[0012] 3-Dimensional NOR memory arrays are disclosed in U.S. Pat. No. 8,630,114 to H. T Lue, entitled “Memory Architecture of 3D NOR Array” , filed on Mar. 11, 2011 and issued on Jan. 14, 2014.

[0013] A three-dimensional memory structure, including horizontal NAND strings that are controlled by vertical polysilicon gates, is disclosed in the article “Multi-layered Vertical gate NAND Flash Overcoming Stacking Limit for Terabit Density Storage” (“Kim”), by W. Kim et al., published in the 2009 Symposium on VLSI Tech. Dig. of Technical Papers, pp 188-189. Another three-dimensional memory structure, also including horizontal NAND strings with vertical polysilicon gates, is disclosed in the article, “A Highly Scalable 8-Layer 3D Vertical-gate (VG) TFT NAND Flash Using Junction-Free Buried Channel BE-SONOS Device,” by H. T. Lue et al., published in the 2010 Symposium on VLSI: Tech. Dig. Of Technical Papers, pp.131-132.

[0014] FIG. 1a shows three-dimensional vertical NAND strings 101 and 102 in the prior art. FIG. 1b shows basic circuit representation 140 of a three-dimensional vertical NAND string in the prior art. Specifically, vertical NAND string 101 and 102 of FIG. 1a and their circuit representation 150 are each essentially a conventional horizontal NAND string which—rather than each connecting 32 or more transistors in series along the surface of a substrate—is rotated 90 degrees, so as to be perpendicular to the substrate. Vertical NAND strings 101 and 102 are serially-connected thin film transistors (TFTs) in a string configuration that rises like a skyscraper from the substrate, with each TFT having a storage element and a control gate provided by one of the word line conductors in an adjacent stack of word line conductors. As shown in FIG. 1b, in the simplest implementation of a vertical NAND string, TFTs 15 and 16 are the first and last memory transistors of NAND string 150, controlled by separate word lines WL0 and WL31, respectively. Bit line select transistor 11, activated by signal BLS, and ground select transistor 12, activated by signal SS, serve to connect an addressed TFT in vertical NAND string 150 to corresponding global bit line GBL at terminal 14 and global source line (ground) GSL, at terminal 13, during read, program, program-inhibit and erase operations. Reading or programming the content of any one TFT, (e.g., TFT 17) requires activation of all 32 TFTs in vertical NAND string 150, which exposes each TFT to read-disturb and program-disturb conditions. Such conditions limit the number of TFTs that can be provided in a vertical NAND string to no more than 64 or 128 TFTs. Furthermore, the polysilicon thin films upon which a vertical NAND string is formed have much lower channel mobility—and therefore higher resistivity—than conventional NAND strings formed in a single-crystal silicon substrate, thereby resulting in a low read current relative to the read current of a conventional NAND string.

[0015] U.S. Patent Application Publication 2011/0298013 (“Hwang”), entitled “Vertical Structure Semiconductor Memory Devices And Methods OF Manufacturing The Same,” discloses three-dimensional vertical NAND strings. In its FIG. 4D, Hwang shows a block of three dimensional vertical NAND strings addressed by wrap-around stacked word lines 150 (reproduced herein as FIG. 1c).

[0016] U.S. Pat. No. 5,768,192 to Eitan, entitled “Memory Cell utilizing asymmetrical charge trapping” filed Jul. 23, 1996 and issued Jun. 16, 1998 discloses NROM type memory transistor operation of the type employed in an embodiment of the current invention.

[0017] U.S. Pat. No. 8,026,521 to Zvi Or-Bach et al, entitled “Semiconductor Device and Structure,” filed on Oct. 11, 2010 and issued on Sep. 27, 2011 to Zvi-Or Bach et al discloses a first layer and a second layer of layer-transferred mono-crystallized silicon in which the first and second layers include horizontally oriented transistors. In that structure, the second layer of horizontally oriented transistors overlays the first layer of horizontally oriented transistors, each group of horizontally oriented transistors having side gates.

[0018] Transistors that have a conventional non-volatile memory transistor structure but short retention times may be referred to as “quasi-volatile.” In this context, conventional non-volatile memories have data retention time exceeding tens of years. A planar quasi-volatile memory transistor on single crystal silicon substrate is disclosed in the article “High-Endurance Ultra-Thin Tunnel Oxide in Monos Device Structure for Dynamic Memory Application”, by H. C. Wann and C. Hu, published in IEEE Electron Device letters, Vol. 16, No. 11, Nov. 1995, pp 491-493. A quasi-volatile 3-D NOR array with quasi-volatile memory is disclosed in the U.S. Pat. No. 8,630,114 to H. T Lue, mentioned above.

[0019] According to one embodiment of the present invention, a high density memory structure, referred to as a three-dimensional vertical NOR Flash memory string (“multi-gate vertical NOR string,” or simply “vertical NOR string”). The vertical NOR string includes a number of thin film transistors (“TFTs”) connected in parallel, having a shared source region and a shared drain region each extending generally in a vertical direction. In addition, the vertical NOR string includes multiple horizontal control gates each controlling a respective one of the TFTs in the vertical NOR string. As the TFTs in a vertical NOR string are connected in parallel, a read current in a vertical NOR string is conducted over a much lesser resistance than the read current through a NAND string of a comparable number of TFTs. To read or program any one of the TFTs in a vertical NOR string, only that TFT needs to be activated, all other TFTs in the vertical NOR string can remain non-conducting. Consequently, a vertical NOR string may include many more TFTs (e.g., several hundreds or more), while allowing faster sensing and minimizing program-disturb or read-disturb conditions.

[0020] In one embodiment, the shared drain region of a vertical NOR string is connected to a global bit line (“voltage V_bl”) and the shared source region of the vertical NOR string is connected to a global source line (“voltage V_ss”). Alternatively, in a second embodiment, only the shared drain region is connected to a global bit line biased to a supply voltage, while the shared source region is pre-charged to a voltage determined by a quantity of charge in the shared source region. To perform the pre-charge, one or more dedicated TFTs may be provided to pre-charge the parasitic capacitance C of the shared source region.

[0021] According to one embodiment of the present invention, multi-gate NOR flash thin film transistor string arrays (“multi-gate NOR string arrays”) are organized as arrays of vertical NOR strings running perpendicular to the surface of a silicon substrate. Each multi-gate NOR string array includes a number of vertical active columns arranged in rows, each row extending along a first horizontal direction, with each active column having two vertical heavily-doped polysilicon regions of a first conductivity, which are separated by one or more vertical polysilicon regions that are undoped or lightly doped to a second conductivity. The heavily-doped regions each form a shared source or drain region and, in conjunction with one or more stacks of horizontal conductors each extending orthogonally to the first horizontal direction, the lightly-doped regions each form multiple channel regions. A charge-trapping material forms storage elements, covering at least the channel regions of TFTs in the active column. The horizontal conductive lines in each stack are electrically isolated from each other and form control gates over the storage elements and the channel regions of the active column. In this manner, the multi-gate NOR string array forms a three-dimensional array of storage TFTs.

[0022] In one embodiment, support circuitry is formed in a semiconductor substrate to support multiple multi-gate NOR string arrays formed above the support circuitry and the semiconductor substrate. The support circuitry may include address encoders, address decoders, sense amplifiers, input/output drivers, shift registers, latches, reference cells, power supply lines, bias and reference voltage generators, inverters, NAND, NOR, Exclusive-Or and other logic gates, other memory elements, sequencers and state machines, among others. The multi-gate NOR string arrays may be organized as multiple blocks of circuits, with each blocks having multiple multi-gate NOR string arrays.

[0023] According to embodiments of the present invention, variations in threshold voltages of TFTs within a vertical NOR string may be compensated by providing one or more electrically programmable reference vertical NOR strings in the same or another multi-gate vertical NOR string array. Background leakage currents inherent to a vertical NOR string can be substantially neutralized during a read operation by comparing the results of the TFT being read to that of a TFT that is concurrently read on a programmable reference vertical NOR string. In some embodiments, each TFT of a vertical NOR string is shaped so as to amplify the capacitive coupling between each control gate and its corresponding channel region thereby to enhance tunneling from the channel regions into the charge-trapping material (i.e., the storage element) during programming, and to reduce the charge injection from the control gate to the charge-trapping material during erasing. This favorable capacitive coupling is particularly useful for storing more than one bit in each TFT of a vertical NOR string. In another embodiment, the charge-trapping material of each TFT may have its structure modified to provide a high write/erase cycle endurance, albeit at a lower retention time that requires refreshing of the stored data. However, as the refreshing required of a vertical NOR string array is expected to be much less frequently than in a conventional dynamic random access memory (DRAM), the multi-gate NOR string arrays of the present invention may operate in some DRAM applications. Such use of the vertical NOR strings allows a substantially lower cost-per-bit figure of merit, as compared to conventional DRAMs, and a substantially lower read-latency, as compared to conventional NAND string arrays.

[0024] In another embodiment the vertical NOR string can be programmed, erased and read as NROM/Mirror-bit TFT string.

[0025] Organizing the TFTs as vertical NOR strings—rather than the prior art vertical NAND strings—results in (i) a reduced read-latency that can approach that of a dynamic random access memory (DRAM) array, (ii) reduced sensitivities to read-disturb and program-disturb conditions that are associated with long NAND Flash strings, and (iii) reduced cost per bit, as compared to a NAND Flash string.

[0026] The present invention is better understood upon consideration of the detailed description below, in conjunction with the accompanying drawings.

[0045] FIG. 2 shows conceptualized memory structure 100, which illustrates a 3-dimensional organization of memory cells (or storage elements) provided in vertical NOR strings. In conceptualized memory structure 100, each vertical NOR string includes memory cells that are each controlled by a corresponding horizontal word line, according to one embodiment of the present invention. In conceptualized memory structure 100, each memory cell is formed in deposited thin films provided “vertically”, i.e., along a direction perpendicular to the surface of substrate layer 101. Substrate layer 101 may be, for example, a conventional silicon wafer used for fabricating integrated circuits, familiar to those of ordinary skill in the art. In this detailed description, a Cartesian coordinate system (such as indicated in FIG. 2) is adopted solely for the purpose of facilitating discussion. Under this coordinate system, the surface of substrate layer 101 is considered a plane which is parallel to the X-Y plane. Thus, as used in this description, the term “horizontal” refers to any direction parallel to the X-Y plane, while “vertical” refers to the Z-direction.

[0046] In FIG. 2, each vertical column in the Z direction represents storage elements or TFTs in a vertical NOR string (e.g., vertical NOR string 121). The vertical NOR strings are arranged in a regular manner in rows each extending along the X direction. (Of course, the same arrangement may also be seen alternatively as an arrangement of rows each extending along the Y directions). The storage elements of a vertical NOR string share a vertical local source line and a vertical local bit line (not shown). A stack of horizontal word lines (e.g., WL 123) run along the Y direction, with each word line serving as control gates for corresponding TFTs of vertical NOR strings located adjacent the word line along the Y direction. Global source lines (e.g., GSL 122) and global bit lines (e.g., GBL 124) are provided along the X direction generally running either below the bottom of or on top of conceptualized memory structure 100. Alternatively, signal lines GSL 122 and GBL 124 can both be routed below or both be routed on top of conceptualized memory structure 100, each of these signal lines may be selectively connected by access transistors (not shown) to the local source lines and local bit lines of individual vertical NOR strings. Unlike a vertical NAND string of the prior art, in a vertical NOR string of the present invention, writing or reading any one of its storage elements does not involve activation of any other storage element in the vertical NOR string. As shown in FIG. 2, solely for illustrative purpose, conceptualized memory block 100 is a multi-gate vertical NOR string array consisting of a 4×5 arrangement of vertical NOR strings, with each NOR string typically having 32 or more storage elements and access selection transistors. As a conceptualized structure, memory block 100 is merely an abstraction of certain salient characteristics of a memory structure of the present invention. Although shown in FIG. 2 as a 4×5 arrangement of vertical NOR strings, with each vertical NOR strings having a number of storage elements, a memory structure of the present invention may have any number of vertical NOR strings in each row along either of the X and Y directions, and any number of storage elements in each vertical NOR string. For example, there may be thousands of vertical NOR strings arrayed in rows along both the X and Y directions, with each NOR string having, for example, 2, 4, 8, 16, 32, 64, 128, or more storage elements.

[0047] The number of storage elements in each vertical NOR string of FIG. 2 (e.g., vertical NOR string 121) corresponds to the number of word lines (e.g., WL 123) providing control gates to the vertical NOR string. The word lines are formed as narrow, long metallic strips each extending along the Y direction. The word lines are stacked one on top of each other, and electrically isolated from each other by dielectric insulation layers there-between. The number of word lines in each stack may be any number, but preferably an integer power of 2 (i.e., 2ⁿ, where n is an integer). The selection of a power of 2 for the number of word lines follows a customary practice in conventional memory design. It is customary to access each addressable unit of memory by decoding a binary address. This custom is a matter of preference and need not be followed. For example, within the scope of the present invention, conceptualized memory structure 100 may have M vertical NOR strings along each row in the X and Y directions, with M being a number that is not necessarily 2ⁿ, for any integer n. In some embodiments to be described below, two vertical NOR strings may share a vertical local source line and a vertical local bit line, but their respective storage elements are controlled by two separate word line stacks. This effectively doubles the storage density of the vertical NOR string.

[0048] As conceptualized memory structure 100 of FIG. 2 is provided merely to illustrate an organization of memory cells, it is not drawn to specific scale in any of the X, Y, Z directions.

[0049] FIG. 3a shows a basic circuit representation in a Z-Y plane of vertical NOR string 300 formed in an active column; vertical NOR string 300 represents a three-dimensional arrangement of non-volatile storage TFTs, with each TFT sharing local source line 355 and local bit line 354, according to one embodiment of the current invention. In this detailed description, the term “active region,” “active column” or “active strip” refers to a region, column or strip of one or more semiconductor materials on which an active device (e.g., a transistor or a diode) may be formed. As shown in FIG. 3a, vertical NOR string 300 runs along the Z direction, with TFTs 316 and 317 connected in parallel between vertical local source line 355 and vertical local drain or bit line 354. Bit line 354 and source line 355 are spaced apart, with the region there-between (i.e., body region 356) providing channel regions for the TFTs in the vertical NOR string. Storage elements are formed at the intersections between channel region 356 and each horizontal word line 323p, where p is the index of the word line in the word line stack; in this example, p may take any value between 0 and 31. The word lines extend along the Y direction. Local bit line 354 is connected through bit line access select transistor 311 to horizontal global bit line (GBL) 314, which runs along the X direction and connects local bit line 354 to access bit line supply voltage Y_bl. Local source line 355 is connected through horizontal global source line (GSL) 313 to source supply voltage V_ss. An optional source-select transistor (not shown in FIG. 3a) can be provided to connect between local source line 355 and GSL 313. The optional source-select transistor may be controlled by source decoding circuitry which can be implemented either in the substrate (e.g., semiconductor substrate 101 of FIG. 2) or above the substrate and below memory structure 100, as is known to a person skilled in the art. Body region 356 of the active column may be connected at terminal 331 to substrate bias voltage V_bb. Substrate bias voltage V_bb may be used, for example during an erase operation. The V_bb supply voltage can be applied to an entire multi-gate vertical NOR string array, or decoded so as to be applied selectively to one or more rows of vertical NOR strings. Lines connecting the V_bb supply voltage to body region 356 run preferably along the direction of the word lines.

[0050] FIG. 3b shows a basic circuit representation in a Z-Y plane of vertical NOR string 305 formed in an active column; vertical NOR string 305 represents a three-dimensional arrangement of non-volatile storage TFTs, including (optionally) dedicated pre-charge TFT 370 for momentarily setting a voltage (“V_ss”) on shared local source line 355, which has a parasitic capacitance C, represented by capacitor 360, according to one embodiment of the present invention. Unlike vertical NOR string 300 of FIG. 3a, vertical NOR string 305 does not implement GSL 313, replacing it with pre-charge transistor 370 which pre-charges parasitic capacitor 360, which temporarily holds a voltage of V_ss volts. Under this pre-charging scheme, global source lines (e.g., global source lines 313 of FIG. 3a) and its decoding circuitry are rendered unnecessary, thereby simplifying both the manufacturing process as well as circuit layout, and providing a very tight footprint for each vertical NOR string. FIG. 3c highlights the structure of non-volatile storage TFT 317, which can also be used, in addition to its normal storage function, to perform the pre-charge function of dedicated pre-charge transistor 370. A dynamic read operation for TFT 317 is described below in conjunction with sensing the correct one of several threshold voltages that is programmed into storage element 334 of TFT 317.

[0051] FIG. 4a is a cross section in a Z-Y plane showing side-by-side active columns 431 and 432, each of which may form a vertical NOR string that has a basic circuit representation illustrated in either FIG. 3a or FIG. 3b, according to one embodiment of the present invention. As shown in FIG. 4a, active columns 431 and 432 each include vertical N+doped local source region 455 and vertical N+ doped local drain or bit line region 454, separated by lightly P− doped or undoped channel region 456. P− doped channel region 456, N+ doped local source region 455 and N+ doped local drain or bit line region 454 may be biased to body bias voltage V_bb, source supply voltage V_ss, and bit line voltage V_bl, respectively. In some embodiments of the current invention, use of body bias voltage V_bb is optional, such as when the active strip is sufficiently thin (e.g., 10 nanometers or less). For a sufficiently thin active strip, the active region is readily fully depleted under appropriate voltage on the control gate, such that voltage V_bb may not provide a solid supply voltage to the channel regions of the TFTs along the vertical NOR string. Isolation region 436, which electrically insulates active columns 431 and 432, may be either a dielectric insulator or an air-gap. A vertical stack of word lines 423p, respectively labeled WL0-WL31 (and optionally WL_CHG), provides control gates to the TFTs in the vertical NOR strings formed in active columns 431 and 432. Word line stack 423p is typically formed as long narrow metallic conductors (e.g., tungsten, a silicide or silicide) that extend along the Y direction, electrically isolated from each other by dielectric layers 426, each typically formed out of silicon oxide (e.g., SiO₂) or an air gap. A non-volatile storage element may be formed at the intersection of each word line 423p and each P− doped channel region 456 by providing a charge-trapping material (not shown) between word line 423p and P− doped channel region 456. For example, FIG. 4a indicates by dashed boxes 416 the locations where nonvolatile storage elements (or storage transistors) T₀ to T₃₁ may be formed. Dashed box 470 indicates where a dedicated pre-charge transistor may be formed, which, when momentarily switched on, allows charge to be transferred from common local bit line region 454 to common local source line region 455 when all transistors T₀ to T₃₁ are in their off state.

[0052] FIG. 4b is a cross section in the Z-X plane showing active columns 430R, 430L, 431R and 431L, charge-trapping layers 432 and 434, and word line stacks 423p-L and 423p-R, according to one embodiment of the present invention. Similar to FIG. 4a, each of vertical word line stacks 423p-L and 423p-R in FIG. 4b denotes a stack of long narrow conductors, where p is an index labeling the word lines in stack (e.g., word lines WL0 to WL31). As shown in FIG. 4b, each word line serves as control gates for the nonvolatile TFTs in the vertical NOR strings formed on adjacent active columns 430-L and 431-R on opposite sides of the word line (within region 490). For example, in FIG. 4b, word line WL31 in word line stack 423p-R serves as control gates for both transistor 416L on active column 430L and transistor 416R on active column 431R. Adjacent word line stacks (e.g., word lines stacks 423p-L and 423p-R) are separated by a distance 495, which is the width of a trench formed by etching through successive word line layers, as described below. Active columns 430R and 430L, and their respective charge-trapping layers 432 and 434, are subsequently formed inside the trench etched through the word line layers. Charge-trapping layer 434 is provided interposed between word line stack 423p-R and vertical active columns 431R and 430L. As elaborated below, during programming of transistor 416R, charge injected into charge-trapping layer 434 is trapped in the portion of charge-trapping layer 434 within dash box 480. The trapped charge alters the threshold voltage of TFT 416R, which may be detected by measuring a read current flowing between local source region 455 and local drain region 454 on active column 431R (these regions are shown, e.g., FIG. 4a in the orthogonal cross section of the active column). In some embodiments, pre-charge word line 478 (i.e., WL_CHG) is provided as control gate of pre-charge TFT 470 that is used to charge parasitic capacitance C of local source line 455 (see, capacitor 360 of FIG. 3b and local source line 455 of FIG. 4a) to a ground or source supply voltage V_ss. For expediency, charge-trapping layer 434 also provides a storage element in pre-charge transistor 470, which however is not itself used as a memory transistor. Pre-charging may alternatively be performed using any of memory transistors T₀ to T₃₁ formed on active column 431R. One or more of these memory transistors, in addition to their storage function, can perform the function of the pre-charge transistor. To perform the pre-charge, the voltage on the word line or control gate is temporarily raised to a few volts above its highest programmable threshold voltage, thereby allowing voltage V_ss applied to local bit line 454 to be transferred to local source line 455 (FIG. 4a). Having memory transistors T₀ to T₃₁ perform the pre-charge charge function eliminates the need for separate dedicated pre-charge TFT 470. Care must be taken, however, to avoid unduly disturbing the threshold voltage of such memory TFT when it is performing its pre-charging function.

[0053] Although active columns 430R and 430L are shown in FIG. 4b as two separate active columns separated by an air-gap or dielectric insulation 433, the adjacent vertical N+ local source lines may be implemented by a single shared vertical local source line. Likewise, the vertical N+ local drain or bit lines may be implemented by a single shared vertical local bit line. Such a configuration provides “vertical NOR string pair”. In that configuration, active columns 430L and 430R may be seen as two branches (hence the “pair”) in one active column. The vertical NOR string pair provides double-density storage through charge-trapping layers 432 and 434 interposed between active columns 430R and 430L and word lines stacks 423p-L and 423p-R on opposite sides. In fact, active columns 430R and 430L may be merged into one active string by eliminating the air gap or dielectric insulation 433, yet still achieve the pair of NOR TFT strings implemented at the two opposite faces of the single active column. Such a configuration achieves the same double-density storage, as the TFTs formed in the opposite faces of the active columns are controlled by separate word line stacks and are formed out of separate charge-trapping layers 434 and 432. Maintaining separate thin active columns 430R and 430L (i.e., instead of merging them into one active column) is advantageous because TFTs on each active column are thinner than the merged column, and can therefore more readily be fully depleted under appropriate control gate voltage conditions, thereby to substantially cut down source-drain subthreshold leakage current between vertical source regions 455 and vertical drain regions 454 of the active columns (FIG. 4a). Having ultra-thin (and therefore highly resistive) active columns is possible for even very long vertical NOR strings (e.g., 128 TFTs or longer) because the TFTs in a vertical NOR string are connected in parallel and because only one of the many TFTs is switched on at any one time, in contrast with the high resistance of a NAND TFT string where TFTs in the string are connected in series and must therefore all be switched on to sense any one of TFTs in the string. For example, in a 32-TFT vertical NOR string, to be able to read transistor T₃₀ (FIG. 4a), the channel length of channel region 456 may span just 20 nanometers, as compared to the corresponding channel length of a NAND string, which may be 32 times longer, or 640 nanometers.

[0054] FIG. 4c shows a basic circuit representation in the Z-X plane of vertical NOR string pairs 491 and 492, according to one embodiment of the present invention. As shown in FIG. 4c, vertical NOR strings 451b and 452a share a common word line stack 423p-R, in the manner shown for the vertical NOR strings of active strips 430L and 431R of FIG. 4b. For their respective commonly-connected local bit lines, vertical NOR string pairs 491 and 492 are served by global bit line 414-1 (GBL1) through access select transistor 411 and global bit line 414-2 (GBL2) through access select transistor 414, respectively. For their respective commonly-connected local source lines, vertical NOR string pairs 491 and 492 are served by global source line 413-1 (GSL1) and global source line 413-2 (GSL2), respectively (source line select access transistors can be similarly provided, and are not shown in FIG. 4c). As shown in FIG. 4c, vertical NOR string pair 491 includes vertical NOR strings 451a and 451b that share local source line 455, local bit line 454, and optional body connection 456. Thus, vertical NOR string pair 491 represent the vertical NOR strings formed on active columns 430R and 430L of FIG. 4b. Word line stacks 423p-L and 423p-R (where, in this example, 31≥p≥0) provide control gates for vertical NOR string 451a and vertical NOR string 451b, respectively. The word lines to control gates in the stack are decoded by decoding circuitry formed in the substrate to ensure that appropriate voltages are applied to the addressed TFT (i.e., the activated word line) and to the unaddressed TFTs (i.e., all other non-activated word lines in the string). FIG. 4c illustrates how storage transistors 416L and 416R on active columns 430L and 431R of FIG. 4b are served by the same word line stack 423p-R. Thus, vertical NOR string 451b of vertical NOR string pair 491 and vertical NOR string 452a of vertical string pair 492 correspond to the adjacent vertical NOR strings formed on active columns 430L and 431R of FIG. 4b. Storage transistors of vertical NOR string 451a (e.g., storage transistor 415R) are served by word line stack 423p-L.

[0055] In another embodiment, the hard-wired global source lines 413-1, 413-2 of FIG. 4c are eliminated, to be substituted for by parasitic capacitance C (e.g., capacitor 460 of FIGS. 4c and 360 of FIG. 3c) between shared N+local source line 455—which is common to both vertical NOR strings 451a and 451b—and its numerous associated word lines 423p-L and 423p-R. In a vertical stack of 32 TFTs, each of the 32 word lines contribute their parasitic capacitance to provide total parasitic capacitance C, such that it is sufficiently large to temporarily hold the voltage supplied by pre-charge TFT 470 to provide a virtual source voltage V_ss during the relatively short duration of read or programming operations. In this embodiment, the virtual source voltage temporarily held on capacitor C is provided to local source line 455 from global bit line GBL1 through access transistor 411 and pre-charge transistor 470. Alternatively, dedicated pre-charge transistor 470 can be eliminated, if one or more of the memory TFTs in the vertical NOR sting are used, in addition to their storage function, to pre-charge local source line 455, by bringing its word line voltage momentarily higher than its highest programmed voltage. Using a storage TFT for this purpose, care must be taken, however, to avoid over-programming the storage TFT. Using the virtual V_ss voltage provides the significant advantage of eliminating hard-wired global source lines (e.g., GLS1, GLS2) and their associated decoding circuitry and access transistors, thereby materially simplifying the process flow and design challenges and resulting in a significant more compact vertical NOR string.

[0056] FIG. 5 is a cross section in the Z-Y plane showing connections of vertical NOR string of active column 531 to global bit line 514-1 (GBL1), global source line 507 (GSL1), and common body bias source 506 (V_bb), according to one embodiment of the present invention. As shown in FIG. 5, bit-line access select transistor 511 connects GBL1 with local bit line 554, and buried contact 556 optionally connects a P− body region on the active strip to body bias source 506 (V_bb) in the substrate. Bit-line access select transistor 511 is formed in FIG. 5 above active column 531. However, alternatively, bit-line access select transistor 511 may be formed at the bottom of active column 531 or in substrate 505 (not shown in FIG. 5). In FIG. 5, bit-line access select transistor 511 can for example be formed in an isolated island of an N+/P−/N+ doped polysilicon stack together with access select word line 585. When a sufficiently large voltage is applied to select word line 585, the P− channel is inverted, thereby connecting local bit line 554 to GBL1. Word line 585 runs along the same direction (i.e., the Y direction) as the word lines 523p which serve as control gates to the TFTs of the vertical NOR string. Word line 585 may be formed separately from word lines 523p. In one embodiment, GBL1 runs horizontally along the X direction (i.e., perpendicular to the directions of the word lines), and bit-line access select transistor 511 provides access to local bit line 554, which is the local bit line of merely one of many vertical NOR strings that are served by GBL1. To increase read and program operation efficiency, in a multi-gate NOR string array, thousands of global bit lines may be used to access in parallel the local bit lines of thousands of vertical NOR strings that are accessed by word line 585. In FIG. 5, local source line 555 is connected through contact 557 to global source line 513-1 (GSL1), which may be decoded, for example by decoding circuitry in substrate 505. Alternatively, as described already, the global source line may be eliminated by providing a virtual source voltage V_ss on local bit line 555 and temporarily pre-charging the parasitic capacitor 560 (i.e., parasitic capacitance C) of local source line 555 through TFT 570.

[0057] Support circuitry formed in substrate 505 may include address encoders, address decoders, sense amplifiers, input/output drivers, shift registers, latches, reference cells, power supply lines, bias and reference voltage generators, inverters, NAND, NOR, Exclusive-Or and other logic gates, other memory elements, sequencers and state machines, among others. The multi-gate NOR string arrays may be organized as multiple blocks of circuits, with each block having multiple multi-gate NOR string arrays.

[0058] FIG. 6a is a cross section in the X-Y plane, showing TFT 685 (T_L) of vertical NOR string 451a and TFT 684 (T_R) of vertical NOR string 451b in vertical NOR string pair 491, as discussed above in conjunction with FIG. 4c. As shown in FIG. 6, TFTs 684 and 685 share N+ local source region 655 and N+ local drain or bit line region 654, both regions extending in long narrow pillars along the Z direction. (N+ local source region 655 corresponds to local source line 455 of FIG. 4a, N+ local drain region 654 corresponds to local bit line 454 of FIG. 4a). In this embodiment, P− doped channel regions 656L and 656R form a pair of active strings between local source pillar 655 and local drain pillar 654 and extend along the Z direction, isolated from each other by isolation region 640. Charge-trapping layer 634 is formed between word lines 623p-L (WL31-0) and 623p-R (WL31-1) and the outside of channel regions 656L and 656R respectively. Charge trapping layer 634 may be a transistor gate dielectric material consisting of, for example, a thin film of tunnel dielectric (e.g., silicon dioxide), followed by a thin layer of charge trapping material such as silicon nitride or conductive nanodots embedded in a non-conducting dielectric material, or isolated floating gates, and is capped by a layer of blocking dielectric such as ONO (Oxide-Nitride-Oxide) or a high dielectric constant film such as aluminum oxide or hafnium oxide or some combination of such dielectrics. Source-drain conduction is controlled by word lines 623p-L and 623p-R, respectively, forming control gates on the outside of charge-trapping layer 634. When programming or reading TFT 684 (T_R), TFT 685 (T_L) is turned off by maintaining an appropriate inhibit voltage at word line 623p-L. Similarly, when programming or reading TFT 685 (T_L), TFT 684 (T_R) is turned off by maintaining an appropriate inhibit voltage at word line 623p-R.

[0059] In the embodiment shown in FIG. 6a, word lines 623p-L and 623p-R are contoured to enhance tunneling efficiency into the TFTs 684 and 685 during programming, while reducing reverse- tunneling efficiency during erasing. Specifically, as is known to a person skilled in the art, curvature 675 of channel region 656R amplifies the electric field at the interface between the active channel polysilicon and the tunneling dielectric during programming, while reducing the electric field at the interface between the word line and the blocking dielectric during erasing. This feature is particularly helpful when storing more than one bit per TFT transistor in a multi-level cell (MLC) configuration. Using this technique, 2, 3, or 4 bits or more may be stored in each TFT. In fact, TFTs 684 and 685 may be used as analog storage TFTs with a continuum of stored states. Following a programming sequence (to be discussed below), electrons are trapped in charge-trapping layer 634, as indicated schematically by dashed lines 680. In FIG. 6a, global bit lines 614-1 and 614-2 run perpendicularly to word lines 623p-R and 623p-L, and are provided either above or underneath the vertical NOR strings, corresponding to bit lines 414-1 and 414-2 respectively of FIG. 4c. As discussed above in conjunction with FIG. 2, the word lines may span the entire length of memory block 100 along the X direction, while the global bit lines span the width of memory block 100 along the Y direction. Of importance, in FIG. 6a, word line 623p-R is shared by TFTs 684 and 683 of two vertical NOR strings on opposite sides of word line 623p-R. Accordingly, to allow TFTs 684 and 683 to be read or programmed independently, global bit line 614-1 (GBL1) contacts local drain or bit line region 657-1 (“odd addresses”), while global bit line 614-2 (GBL2) contacts local drain or bit line region 657-2 (“even addresses”). To achieve this effect, contacts along global bit lines 614-1 and 614-2 are staggered, with each global bit line contacting every other one of the vertical NOR string pair along the X-direction row.

[0060] In like manner, global source lines (not shown in FIG. 6a), which may be located either at the bottom or above the multi-gate NOR string array, may run parallel to the global bit lines and may contact the local source lines of vertical NOR string pairs according to even or odd addresses. Alternatively, where pre-charging of parasitic capacitance C (i.e., capacitor 660) temporarily to virtual source voltage V_ss is used, the global source lines need not be provided, thereby simplifying the decoding scheme as well as the process complexity.

[0061] FIG. 6a shows only one of several possible embodiments by which vertical NOR string pairs may be provided with stacked word lines. For example, curvature 675 in channel region 656R can be further accentuated. Conversely such curvature can be altogether eliminated (i.e. straightened out) as shown in the embodiment of FIG. 6b. In the embodiment of FIG. 6b isolation spacing 640 of FIG. 6a may be reduced or altogether eliminated by merging channel regions 656L and 656R into a single region 656(L+R), achieving greater area efficiency without sacrificing the dual-channel configuration: for example TFTs 685 (T_L) and 684 (T_R) reside on opposite faces of the same active strip. In the embodiments of FIGS. 6a, 6b, vertical NOR strings sharing a word line may be laid out in a staggered pattern relative to each other (not shown), such that they may be brought closer to each other, so as to reduce the effective footprint of each vertical NOR string. Although FIGS. 6a and 6b show direct connection via a contact between global bit line 614-1 and N+ doped local drain bit line pillar 654 (LBL-1), such connection can also be accomplished using a bit-line access selection transistor (e.g., bit line access select transistor 511 of FIG. 5, not shown in already crowded FIGS. 6a and 6b).

[0062] In the embodiments of FIGS. 6a and 6b, dielectric isolation between N+ doped local drain region 654 and its adjacent local N+ doped source region 658 (corresponding to isolation region 436 of FIG. 4a) can be established by, for example, defining the separation 676 between word lines 623p-R and 623p-L to be less than the thicknesses of two back-to-back charge-trapping layers, so that the charge-trapping layers are merged together during their deposition. The resulting merging of the deposited charge-trapping layers creates the desired dielectric isolation. Alternatively, isolation between adjacent active strings can be achieved by using a high aspect-ratio etch of N+ polysilicon to create gap 676 (air gap or dielectric filled) isolating N+ pillar 658 of one string from N+ pillar 654 of the adjacent string (i.e., creating gap 436 shown in FIG. 4a).

[0063] Contrasting between the prior art vertical NAND strings and the vertical NOR strings of the current invention, although both types of devices employ thin film transistors with similar word line stacks as control gates, their transistor orientations are different: in the prior art NAND string, each vertical active strip may have 32, 48 or more TFTs connected in series. In contrast, each active column forming the vertical NOR strings of the present invention the vertical column may have one or two sets of 32, 48 or more TFTs connected in parallel. In the prior art NAND strings, the word lines in some embodiments typically wrap around the active strip. In some embodiments of the vertical NOR string of the present invention separate designated left and right word lines are employed for each active strip, thereby to achieve a doubling (i.e. a pair) storage density for each global bit line, as illustrated in FIGS. 4c, 6a and 6b. The vertical NOR strings of the present invention do not suffer from program-disturb or read-disturb degradation, nor do they suffer from the slow latency of the prior art NAND strings. Thus, a much larger number of TFTs may be provided in a vertical NOR string than in a vertical NAND strings. Vertical NOR strings, however, may be more susceptible to subthreshold or other leakage between the long vertical source and drain diffusions (e.g., local source region 455 and local drain region 454, respectively, illustrated in FIG. 4a).

[0064] Two additional embodiments of the vertical NOR string of this invention are shown in FIG. 6c and FIG. 6d. In these embodiments, all word lines in each word-line stack wrap around the vertical active strip.

[0065] In FIG. 6c, a vertical NOR string is formed inside the voids that are formed by etching through a stack of metal word lines and the dielectric isolation layers between the word lines. The manufacturing process flow is similar to that of the prior art vertical NAND strings, except that the transistors in a vertical NOR string are provided parallel to each other, rather than serially in a vertical NAND string. Formation of transistors in a vertical NOR string is facilitated by the N+ doped vertical pillars extending to the entire depth of the void, providing shared local source line 655 (LSL) and shared local bit line (drain) 654 (LBL) for all the TFTs along the vertical NOR string, with undoped or lightly doped channel region 656 adjacent to both. Charge storage element 634 is positioned between channel 656 and word line stack 623p, thus forming a stack of 2, 4, 8, . . . 32, 64 or more TFTs (e.g., device 685 (T₁₀)) along the vertical active strip. In the embodiment of FIG. 6c, the word line stacks run in the Y direction, with individual horizontal strips 623p (WL31-0), 623p (WL31-1) being separated from each other by air gap or dielectric isolation 610. Global bit lines 614 (GBL) and global source lines 615 (GSL) run horizontally in rows along the X direction, perpendicular to the word lines. Each global bit line 614 accesses local bit line pillars 654 (LBL) along the row of vertical strips through access select transistors (511 in FIG. 5, not shown here) that can be positioned either below the memory array or above it. Similarly, each global source line 615 accesses the local source line pillars along the row. While the structures shown in FIGS. 6a and 6b are able to fit a pair of vertical NOR strings in roughly the same area taken up by a single vertical NOR string in the embodiment of FIG. 6c, each TFT in each vertical NOR string shown in FIG. 6c has two parallel conduction channels (i.e., channel regions 656a and 656b), and therefore may store more charge and increase or double the read current, thereby enabling storing more bits in each TFT.

[0066] FIG. 6d shows a more compact vertical NOR string with wrap-around word lines, according to one embodiment of the present invention. As shown in FIG. 6d, vertical NOR strings are staggered as to be closer together, so that word line stack 623p (WL31-0) can be shared by more vertical NOR strings. The staggered configuration is enabled by the use of parasitic capacitance C (i.e., capacitors 660) of local source line pillar 655 (LSL). By pre-charging capacitors 660 to temporarily hold virtual voltage V_ss during read and program operations, as described below, the need for hard-wired global source lines (e.g., GSL 615 in FIG. 6c) is obviated. Although the vertical NOR strings of FIGS. 6c and 6d may not by themselves offer significant areal efficiencies, as compared to prior art vertical NAND strings (e.g., the NAND strings of FIG. 1c), such vertical NOR strings achieve much greater string lengths than vertical NAND strings. For example, while vertical NOR strings of the present invention may well support strings of length 128 to 512 or more TFTs in each stack, such string lengths are simply not practical for a vertical NAND string, given the serious limitations attendant with series-connected TFT strings.

Fabrication Process

[0067] FIGS. 7a, 7b, 7c and 7d are cross sections of intermediate structures formed in a fabrication process for a multi-gate NOR string array, in accordance with one embodiment of the present invention.

[0068] FIG. 7a shows a cross section in the Z-Y plane of semiconductor structure 700, after low resistivity layers 723p have been formed above substrate 701, in accordance with one embodiment of the present invention. In this example, p is an integer between 0 and 31, representing each of 32 word lines. As shown in FIG. 7a, semiconductor structure 700 includes low resistivity layers 723-0 to 723-31. Semiconductor substrate 701 represents, for example, a P− doped bulk silicon wafer on and in which support circuits for memory structure 700 may be formed prior to forming the vertical NOR strings. Such support circuits may include both analog and digital logic circuits. Some examples of such support circuits may include shift registers, latches, sense amplifiers, reference cells, power supply lines, bias and reference voltage generators, inverters, NAND, NOR, Exclusive-OR and other logic gates, input/output drivers, address decoders, including bit-line and word line decoders, other memory elements, sequencers and state machines. To provide these support circuits, the building blocks of conventional N-Wells, P-Wells, triple wells (not shown), N⁺ diffusion regions (e.g., region 707-0) and P⁺ diffusion regions (e.g., region 706), isolation regions, low and high voltage transistors, capacitors, resistors, diodes and interconnects are provided, as known to a person skilled in the art.

[0069] After the support circuits have been formed in and on semiconductor substrate 701, insulating layers 708 are provided, which may be deposited or grown thick silicon dioxide, for example. In some embodiments, one or more metallic interconnect layers may be formed, including global source line 713-0, which may be provided as horizontal long narrow strips running along a predetermined direction. Global source line 713-0 is connected through etched openings 714 to circuitry 707 in substrate 701. To facilitate discussion in this detailed description, the global source lines are presumed to run along the X direction. The metallic interconnect lines may be formed by applying photo-lithographical patterning and etching steps on one or more deposited metal layers. (Alternatively, these metallic interconnect lines can be formed using a conventional damascene process, such as a conventional copper or tungsten damascene process). Thick dielectric layer 709 is then deposited, followed by planarization using conventional chemical mechanical polishing (CMP).

[0070] Conductor layers 723-0 to 723-31 are then successively formed, each conductor layer being insulated from the layer underneath it and the layer above it by an intervening insulating layers 726. In FIG. 7a, although thirty two conductor layers are indicated, any number of such layers may be provided. In practice, the number of conductor layers that can be provided may depend on the process technology, such as the availability of a well-controlled anisotropic etching process that allows cutting through the multiple conductor layers and dielectric isolation layers 726 there-between. For example, conductor layers 723p may be formed by first depositing 1-2 nm thick layer of titanium nitride (TiN), followed by depositing a 10-50 nm thick layer of tungsten (W) or a similar refractory metal, or a silicide such as silicides of nickel, cobalt or tungsten among others, or a salicide, followed by a thin layer of etch-stop material such as aluminum oxide (Al₂O₃). Each conductor layer is etched in a block 700 after deposition, or is deposited as a block through a conventional damascene process. In the embodiment shown in FIG. 7a, each successive conductor layer 723p extends in the Y-direction a distance 727 short of (i.e. recessed from) the edge of the immediately preceding metal layer, so that all conductor layers may be contacted from the top of structure 700 at a later step in the process. However, to reduce the number of masking and etch steps necessary to form the stepped conductors stack of FIG. 7a, it is possible to achieve recessed surfaces 727 simultaneously for multiple conductor layers by employing other process techniques known to a person skilled in the art that do not require each individual conductor plane to be separately masked and etched to create exposed recessed surfaces 727. After the conductor layer is deposited and etched, the corresponding one of dielectric isolation layers 726 is then deposited. Dielectric isolation layers 726 may be, for example, a silicon dioxide of a thickness between 15 and 50 nanometers. Conventional CMP prepares the surface of each dielectric layer for depositing the next conductor layer. The number of conductor layers in the stack of block 700 corresponds to at least the number of memory TFTs in a vertical NOR string, plus any additional conductor layers that may be used as control gates of non-memory TFTs such as pre-charge TFTs (e.g., pre-charge TFT 575 of FIG. 5), or as control gates of bit-line access select TFTs (e.g., 585 bit-line access select TFT 511 of FIG. 5). The conductor layer deposition and etch steps and the dielectric layer deposition and CMP process are repeated until all conductor layers are provided.

[0071] Dielectric isolation layer 710 and hard mask layer 715 are then deposited. Hard mask 715 is patterned to allow etching of conductor layers 723p to form long strips of yet to be formed word lines. The word lines extend in length along the Y direction. One example of a masking pattern is shown in FIG. 6 for word lines 623p-R, 623p-L, which includes features such as the extensions in adjacent word lines towards each other at separation 676 and the recesses in each word line to create the desired curvatures 675. Deep trenches are created by anisotropically etching through successive conductor layers 723p and their respective intervening dielectric insulator layers 726, until dielectric layer 709 at the bottom of conductor layers 723p is reached. As a large number of conductor layers are etched, a photoresist mask by itself may not be sufficiently robust to hold the desired word line pattern through numerous successive etches. To provide a robust mask, hard mask layer 715 (e.g., carbon) is preferred, as is known to a person of ordinary skill in the art. Etching may terminate at dielectric material 709, or at landing pads 713 on the global source lines, or at substrate 701. It may be advantageous to provide an etch-stop barrier film (e.g., aluminum oxide) to protect landing pads 713 from etching.

[0072] FIG. 7b illustrates, in a cross section in the Z-X plane of semiconductor structure 700, etching through successive conductor layers 723p and corresponding dielectric layers 726 to form trenches (e.g., deep trench 795), which reach down to dielectric layer 709, according to one embodiment of the present invention. In FIG. 7b, conductor layers 723p are anisotropically etched to form conductor stacks 723p-R and 723p-L, which are separated from each other by deep trench 795. This anisotropic etch is a high aspect-ratio etch. To achieve the best result, etch chemistry may have to be alternated between conductor material etch and dielectric etch, as the materials of the different layers are etched through, as in known to a person skilled in the art. The anisotropy of the multi-step etch is important, as undercutting of any of the layers should be avoided, so that a resulting word line at the bottom of a stack would have approximately the same conductor width and trench spacing as the corresponding width and spacing of a word line near or at the top of the stack. Naturally, the greater the number of conductor layers in the stack, the more challenging it becomes to maintain a tight pattern tolerance through the numerous successive etches. To alleviate the difficulty associated with etching through, for example, 64 or 128 or more conductor layers, etching may be conducted in sections of, say, 32 layers each. The separately etched sections can then be stitched together, as taught, for example, in the Kim reference mentioned above.

[0073] Etching through multiple conductor layers 723p of conductor material (e.g., tungsten or other refractory materials) is much more difficult and time-consuming than etching of the intervening insulating layers 726. For that reason, an alternative process may be adopted that eliminates the need for multiple etches of conductor layers 723p. That process, well known to a person skilled in the art, consists of first substituting sacrificial layers of a readily etchable material in place of conductor layers 723p of FIG. 7b. For example, insulating layers 726 can be silicon dioxide and sacrificial layers (occupying the spaces shown as 723p in FIG. 7b) can be silicon nitride or another fast etching dielectric material. Deep trenches are then etched anisotropically through the ONON (Oxide-Nitride-Oxide-Nitride) alternating dielectric layers to create tall stacks of the dual dielectrics. At a later step in the manufacturing process flow (to be described below), these stacks are supported by active vertical strips of polysilicon, allowing the sacrificial layers to be etched away, preferably through selective chemical or isotropic etch. The cavities thus created are then filled through conformal deposition of the conductor material, resulting in conductor layers 723p separated by intervening insulating layers 726.

[0074] After the structure of FIG. 7b is formed, charge-trapping layers 734 and polysilicon layers 730 are then deposited in succession conformally on the vertical sidewalls of the etched conductor word line stacks. A cross section in the Z-X plane of the resulting structure is shown in FIG. 7c. As shown in FIG. 7c, charge-trapping layers 734 are formed, for example, by first depositing blocking dielectric 732a, between 5 to 15 nanometers thick and consisting of a dielectric film of a high dielectric constant (e.g., aluminum oxide, hafnium oxide, or some combination silicon dioxide and silicon nitride). Thereafter, charge-trapping material 732b is deposited to a thickness of 4 to 10 nanometers. Charge-trapping material 732b may be, for example, silicon nitride, silicon-rich oxynitride, conductive nanodots embedded in a dielectric film, or thin conductive floating gates isolated from adjacent TFTs sharing the same vertical active strip. Charge-trapping 732b may then be capped by a deposited conformal thin tunnel dielectric film in the thickness range of 2 to 10 nanometers (e.g., a silicon dioxide layer, or a silicon oxide-silicon nitride-silicon oxide (“ONO”) triple-layer). The storage element formed out of charge-trapping layers 734 may be any one of SONOS, TANOS, nanodot storage, isolated floating gates or any suitable charge-trapping sandwich structures known to a person of ordinary skill in the art. The combined thickness of charge-trapping layers 734 is typically between 15 and 25 nanometers.

[0075] After deposition of charge-trapping layer 734, contact openings are made at the bottom of trench 795, using a masking step and by anisotropically etching through charge-trapping layers 734 and dielectric layer 709 at the bottom of trench 795, stopping at bottom global source line landing pad 713 for the source supply voltage V_ss (see, FIG. 7b), or at global bit line voltage V_bl (not shown), or at P+ region 706 for contact to a back bias supply voltage V_bb (see, FIG. 7c). In some embodiments, this etch step is preceded by a deposition of an ultra-thin film of polysilicon (e.g. 2 to 5 nanometers thick) to protect the vertical surfaces of tunnel dielectric layer 732c during the contact-opening etch of charge-trapping material 734 at the bottom of trench 795. In one embodiment, each global source line is connected only to alternate ones in a row of vertical NOR string pairs. For example, in FIG. 5, for odd address word lines, electrical contacts (e.g., contact opening 557) are etched to connect the N+ doped local source lines (e.g., local source line 555 in FIG. 5) to global source line 513-1. Likewise, for even address word lines, electrical contacts are etched to connect the N+ doped local source lines in the row of vertical NOR string pairs to global source line 513-2 (not shown in FIG. 5). In the embodiment employing virtual V_ss through parasitic capacitor C (i.e., capacitors 560 in FIG. 5) the step of etching through charge trapping layer 734 at the bottom of trench 795 may be skipped.

[0076] Thereafter, polysilicon thin film 730 is deposited to a thickness ranging between 5 and 10 nanometers. In FIG. 7c, polysilicon thin film 730 is shown on the opposite sidewalls of trench 795, labeled respectively 730R and 730L. Polysilicon thin film 730 is undoped or preferably doped P− with boron, at a doping concentration typically in the range of 1×10¹⁶ per cm³ to 1×10¹⁷ per cm³, which allows a TFT to be formed therein to have an enhancement native threshold voltage. Trench 795 is sufficiently wide to accommodate charge-trapping layers 734 and polysilicon thin film 730 on its opposing sidewalls. Following the deposition of polysilicon 730, the sacrificial layers in the stack described above are etched away and the cavities thus formed are filled with the conformally deposited conductor layers 723p (FIG. 7c).

[0077] As shown in FIG. 7b, trench 795 extends along the Y direction. After formation of isolated word line stacks 723p-L and 723p-R, in one example semiconductor structure 700 may have 16,000 or more side-by-side word line stacks, each serving as control gates for 8,000 or more active columns to be formed along the length of each stack, or 16,000 TFTs (8,000 TFTs on each side of the stack). With 64 word lines in each stack, 16 billion TFTs may eventually be formed in each of such multi-gate vertical NOR string array. If each TFT stores two data bits, such a multi-gate vertical NOR string array would store 32 gigabits of data. Approximately 32 such multi-gate vertical NOR string arrays (plus spare arrays) may be formed on a single semiconductor substrate, thereby providing a 1-terabit integrated circuit chip.

[0078] FIG. 7d is a cross section view in the X-Y plane of the top surface of the structure of FIG. 7c in one embodiment. Nestled between word lines 723p-L and 723p-R are the two sidewalls 730L and 730R of the vertical deposited P− doped polysilicon structure (i.e., an active column). The deep void 740 between sidewalls 730L and 730R may be filled with a fast-etching insulating dielectric material (e.g., silicon dioxide or liquid glass or carbon doped silicon oxide). The top surface may then be planarized using conventional CMP. A photolithographic step then exposes openings 776 and 777, which is followed by a high aspect-ratio selective etching to excavate the fast-etching dielectric material in exposed areas 776 and 777 all the way down to the bottom of trench 795. A hard mask may be required in this etching step to avoid excessive pattern degradation during etch. The excavated voids are then filled with an in-situ N+ doped polysilicon. The N+ dopants diffuse into the very thin lightly doped active polysilicon pillars 730L and 730R within the exposed voids to make them N+ doped. Alternatively, prior to filling the voids with the in-situ N+ doped polysilicon the lightly doped polysilicon inside the voids can be etched away through a brief isotropic plasma etch or selective wet etch. CMP or top surface etching then removes the N+ polysilicon from the top surface, leaving tall N+ polysilicon pylons in areas 754 (N+) and 755(N+). These N+ pylons form the shared vertical local source line and the shared vertical local bit line for the TFTs in the resulting vertical NOR strings.

[0079] Next, a dielectric isolation layer is deposited and patterned using photolithographic masking and etching steps. The etching step opens contacts for connecting the vertical local bit lines to the horizontal global bit lines (e.g., contacts 657-1 to strings at odd addresses and 657-2 to strings at even addresses, as shown in FIG. 6). A low resistivity metal layer (e.g., tungsten) is deposited. The deposited metal is then patterned using photolithographic and etching steps to form global bit-lines (e.g., global word line 614-1 or GBL1 for strings at odd addresses, and global bit line 614-2 (GBL2) for strings at even addresses, as shown in FIG. 6). Alternatively, the global bit lines may be formed using conventional copper damascene process. All global bit lines, as well as all metal layers 723p of the word line stacks (FIG. 7a) are connected by etched vias to word line and bit-line decoding and sensing circuits in the substrate, as is known to a person skilled in the art. Switch and sensing circuits, decoders and reference voltage sources can be provided to global bit lines and global word lines, either individually or shared by several ones of the bit lines and word lines.

[0080] In some embodiments, bit line access select transistors (511 in FIG. 5) and their associated control gate word lines (e.g., word lines 585 in FIG. 5) are formed as isolated vertical N+P−N+ transistors, as known to a person skilled in the art, to selectively connect odd and even global bit lines (e.g., bit lines 614-1 and 614-2 in FIG. 6a) to vertical NOR strings at alternate odd and even addresses (e.g., local bit lines 657-1 and 657-2, respectively, in FIG. 6a).

Read Operation

[0081] Because the TFTs of a vertical NOR string are connected in parallel, in all embodiments of the current invention, all TFTs in an active column (including an active column having formed thereon a vertical NOR string pair) should preferably be in enhancement mode—i.e., each TFT should have a positive gate-to-source threshold voltage—so as to suppress leakage currents during a read operation between the shared local source line and the shared local bit line (e.g., local bit line 455 and local source line 454 shown in FIG. 4c). Enhancement mode TFTs are achieved by doping the channel regions (e.g., P− channel region 756 of FIG. 7c) with boron in a concentration typically between 1>10¹⁶ and 1>10¹⁷ per cm³, targeting for a native TFT threshold voltage of around 1V. With such TFTs, all unselected word lines in the vertical NOR string pair of an active column may be held at 0V. Alternatively, the read operation may raise the voltage on the shared local N+ source line (e.g., local source line 455 of FIG. 4c) to around 1.5V, while raising the voltage on the shared local N+ drain line (e.g., local bit line 454) to around 2V and holding all unselected local word lines at 0V. Such a configuration is equivalent to setting the word line to −1.5V with respect to the source, thereby suppressing leakage current due to TFTs that are in slightly depleted threshold voltage, which occurs, for example, if the TFTs are slightly over-erased.

[0082] After erasing the TFTs of a vertical NOR string, a soft programming operation may be required to shift any TFT in the vertical NOR string that is over-erased (i.e., now having a depletion mode threshold voltage) back to an enhancement mode threshold voltage. In FIG. 5, an optional connection 556 is shown by which P− channel is connected to back bias voltage 506 (V_bb,) (also shown as body connection 456 in FIG. 4c). A negative voltage may be used for V_bb to modulate the threshold voltage of the TFTs in each active column to reduce subthreshold leakage currents between the shared N+ source and the shared N+ drain/local bit line. In some embodiments, a positive V_bb voltage can be used during an erase operation to tunnel-erase TFTs whose control gates are held at 0V.

[0083] To read the data stored in a TFT of a vertical NOR string pair, all TFTs on both vertical NOR strings of the vertical NOR string pair are initially placed in the “off” state by holding all word lines in the multi-gate NOR string array at 0V. The addressed vertical NOR string can either share a sensing circuit among several vertical NOR strings along a common word line through use of decoding circuitry. Alternatively, each vertical NOR string may be directly connected through a global bit-line (e.g., GBL1 of FIG. 4c) to a dedicated sensing circuit. In the latter case, one or more vertical NOR strings sharing the same word line plane may be sensed in parallel. Each addressed vertical NOR string has its local source line set at V_ss˜0V, either through its hard-wired global source line (e.g., GSL1 in FIG. 4c) as shown schematically in FIG. 8a, or as a virtual V_ss˜0V through a pre-charge transistor (e.g., pre-charge transistor 470 in FIG. 4c or transistor 317 in FIG. 3c) which momentarily transfers V_bl˜0V to parasitic capacitance C (e.g., capacitor 460 or capacitor 360) of floating local source line 455 or 355) during the pre-charge, as shown schematically in FIG. 8b.

[0084] Immediately after turning off pre-charge transistor 470, the local bit line (e.g., local bit line 454 of FIG. 4c) is set at V_bl˜2V through the bit line access select transistor (e.g., bit line access select transistor 411 of FIG. 4c or access select transistor 511 in FIG. 5). V_bl˜2V is also the voltage at the sense amplifiers for the addressed vertical NOR strings. At this time, the addressed word line is raised in small incremental voltage steps from 0V to typically about 6V, while all the un-selected word lines at both the odd address TFTs and the even address TFTs of the vertical NOR string pair remain at 0V. In the embodiment of hard-wired V_ss of FIG. 8a, the addressed TFT has been programmed in one example to a threshold voltage of 2.5V, therefore the voltage V_bl at local bit line LBL begins to discharge through the selected TFT towards the OV of the local source line (V_ss) as soon as its WL_s exceeds 2.5V, thus providing a voltage drop (shown by the dashed arrow in FIG. 8a) that is detected at the sense amplifier serving the selected global bit line. In the embodiment of the virtual V_ss of FIG. 8b, pre-charge transistor word line WL_CHG momentarily is turned on to pre-charge floating local source line LSL to 0V at the start of the read sequence. Then, selected word line WL_s goes through its incremental voltage steps, and as soon as it exceeds the programmed 2.5V, the selected TFT momentarily dips the voltage on its local bit line from its V_bl˜2V. This voltage dip (shown by the dashed arrow in FIG. 8b) is detected by the sense amplifier of the global bit line connected to the selected local bit line. There are other alternative schemes to correctly read the programmed threshold voltage of the selected TFT as known to a person skilled in the art. It should be pointed out that for the embodiments relying on parasitic capacitance C to temporarily hold virtual voltage V_ss, the higher the vertical stack the bigger is capacitance C and therefore the longer is the hold time and the greater is the read signal presented to the selected sense amplifier. To further increase C it is possible to add in one embodiment one or more dummy conductors in the vertical string whose primary purpose is to increase capacitance C.

[0085] In the case of an MLC implementation (i.e., a “multi-level cell” implementation, in which each TFT stores more than one bit), the addressed TFT may have been programmed to one of several voltages (e.g., 1V (erased state), 2.5V, 4V or 5.5V). The addressed word line WLs is raised in incremental voltage steps until conduction in the TFT is detected at the sense amplifier. Alternatively, a single word line voltage can be applied (e.g., ˜6 volts), and the rate of discharge of the local bit line LBL (V_bl) can be compared with the rates of discharge from several programmable reference voltages representative of the voltage states of the stored multi-bit. This approach can be extended for a continuum of states, effectively providing analog storage. The programmable reference voltages maybe stored in dedicated reference vertical NOR strings located within the multi-gate vertical NOR string array, so that the characteristics during read, program, and background leakage are closely tracked. In a vertical NOR string pair, only the TFTs on one of the two vertical NOR strings can be read in each read cycle; the TFTs on the other vertical NOR string are placed in the “off” state (i.e., all word lines at 0V). During a read cycle, as only one of the TFTs in a vertical NOR string is exposed to the read voltages, read disturb conditions are essentially absent.

[0086] In one example of an embodiment of this invention, 64 TFTs and one or more pre-charge TFTs may be provided on each vertical NOR string of a vertical NOR string pair. Each word line at its intersection with the local vertical N+ source line pillar forms a capacitor (see, e.g., capacitor 660 of FIG. 6a). A typical value for such a capacitor may be, for example, 1×10⁻¹⁸ farads. Including all the capacitors in both vertical NOR strings of a vertical NOR string pair, the overall distributed capacitance C totals approximately 1×10⁻¹⁶ farads, which is sufficient for a local source line to preserve a pre-charged source voltage (V_ss) during a read cycle, which is completed in typically less than a microsecond immediately following the pre-charge operation. The charging time through bit-line access select transistors 411 and pre-charge TFT 470 is in the order of a few nanoseconds, thus the charging time does not add noticeably to the read latency. Reading from a TFT in a vertical NOR string is fast, as the read operation involves conduction in only one of the TFTs in the vertical NOR string, unlike the read operation on a NAND string, in which many TFTs connected in series are required to be conducting.

[0087] There are two major factors contributing to the read latency of vertical NOR strings of the current invention: (a) the RC time delay associated with resistance R_bl and capacitance C_bl of a global bit line (e.g., GBL 614-1 in FIG. 6a), and (b) the response time of a sense amplifier to a voltage drop V_bl on the local bit line (e.g., LBL-1) when the addressed TFT begins conducting. The RC time delay associated with a global bit line serving, for example, 16,000 vertical NOR strings is of the order of a few tens of nanoseconds. The read latency for reading a TFT of a prior art vertical NAND string (e.g., the NAND string of FIG. 1b) is determined by the current through 32 or more series-connected TFTs and select transistors discharging capacitance C_bl of the global bit line. By contrast, in a vertical NOR string of the present invention, the read current discharging C_bl is provided through just the one addressed transistor (e.g., transistor 416L of FIG. 4a) in series with bit line access select transistor 411, resulting in a much faster discharge of the local bit line voltage (V_bl). As a result, a much lower latency is achieved.

[0088] In FIG. 4c, when one TFT (e.g., TFT 416L in the vertical NOR string 451b) is read at a time, all other TFTs in either vertical NOR string 451a and 451b of vertical NOR string pair 491 are held in their “off” states, their word lines being held at 0V. Even though TFT 416R in vertical NOR string 452a of vertical NOR string pair 492 shares word line W31 with TFT 416L, TFT 416R may be read simultaneously with TFT 416L because vertical NOR string 452a is served by global bit line 414-2, while vertical NOR string 451b is served by global bit line 414-1. (FIGS. 6a and 6b illustrate how global bit lines 614-1 and 614-2 serve adjacent vertical NOR string pairs).

[0089] In one embodiment, a word line stack includes 32 or more word lines provided in 32 planes. In one multi-gate vertical NOR string array, each plane may include 8000 word lines controlling 16,000 TFTs, each of which may be read in parallel through 16,000 global bit lines, provided that each bit line is connected to a dedicated sense amplifier. Alternatively, if several global bit lines share a sense amplifier through a decode circuit, the 16000 TFTs are read over several successive read cycles. Reading in parallel a massive number of discharging TFTs can cause a voltage bounce in the ground supply (V_ss) of the chip, which may result in read errors. However, an embodiment that uses the pre-charged parasitic capacitor C in the local source line (i.e., providing a virtual source voltage (V_ss) for vertical NOR string) has a particular advantage in that such ground voltage bounce is eliminated. This is because the virtual source voltages in the vertical NOR strings are independent and are not connected to the ground supply of the chip.

Program (Write) and Program-Inhibit Operations.

[0090] Programming of an addressed TFT may be achieved by tunneling—either direct tunneling or Fowler-Nordheim tunneling,—of electrons from the channel region of the TFT (e.g., channel region 430L shown in FIG. 4b) to the charge-trapping layer (e.g., charge trapping layer 434) when a high programming voltage is applied between the selected word line (e.g., word line 423p-R) and the active channel region (e.g., active channel region 456 in FIG. 4a). As tunneling is highly efficient, requiring very little current to program a TFT, parallel programming of tens of thousands of TFTs may be achieved at low power dissipation. Programming by tunneling may require, for example, a 20V, 100-microsecond pulse. Preferably, the programming is implemented through a succession of shorter duration stepped voltage pulses, starting at around 14V and going as high as approximately 20V. Stepped voltage pulsing reduces electrical stress across the TFT and avoids overshooting the intended programmed threshold voltage.

[0091] After each programming high-voltage pulse the addressed transistor is read to check if it has reached its target threshold voltage. If the target threshold voltage has not been reached, the next programming pulse applied to the selected word line is incremented typically by a few hundred millivolts. This program-verify sequence is repeatedly applied to the one addressed word line (i.e., a control gate) with 0V applied to the local bit line (e.g., local bit line 454 of FIG. 4a) of the active column (e.g., column 430L of FIG. 4b). At these programming high word line voltages, TFT 416L's channel region is inverted and is held at 0V, so that electrons tunnel into the charge storage layer of TFT 416L. When the read sensing indicates that the addressed TFT has reached its target threshold voltage, the addressed TFT must be inhibited from further programming, while other TFTs sharing the same word line may continue programming to their higher target threshold voltages. For example, when programming TFT 416L in vertical NOR string 451b, programming of all other TFTs in vertical NOR strings 451b and 451a must be inhibited by keeping all their word lines at 0V.

[0092] To inhibit further programming or TFT 416L once it has reached its target threshold voltage, a half-select voltage (i.e., approximately 10V) is applied to local bit line 454. With 10V being placed in the channel region and 20V being placed on the control gate, only net 10V is applied across the charge trapping layer, therefore the Fowler-Nordheim tunneling current is insignificant and no meaningful further programming takes place on TFT 416L during the remaining sequence of stepped pulse voltages up to the maximum 20V. By raising the local bit line 454 to 10V while continuing to increment the programming voltage pulses on word line WL31, all TFTs on vertical NOR strings sharing the same selected word line are programmed correctly to their higher target threshold voltages. The sequence of “program-read-program inhibit” is indispensable for correctly programming tens of thousands TFTs in parallel to their various target threshold voltage states in multilevel cell storage. Absent such program inhibit of individual TFTs over-programming may cause overstepping or merging with the threshold voltage of the next higher target threshold voltage state. Although TFT 416R and TFT 416L share the same word line, they belong to different vertical NOR string pairs 452 and 451. It is possible to program both TFT 416L and TFT 416R in the same programming pulsed voltage sequence, as their respective bit line voltages are supplied through GBL1 and GBL2 and are independently controlled. For example, TFT 416L can continue to be programmed while TFT 416R can be inhibited from further programming at any time. These program and program-inhibit voltage conditions can be met because vertical NOR strings 451a and 451b of vertical NOR string pair 491 are controlled by separate word lines 423p-L and 423p-R respectively, and the voltage on each local bit line can be set independently from all other vertical NOR string pairs. During programming, any unselected word line within an addressed word line stack or within unaddressed word line stacks can be brought to 0V, half-select 10 volts, or floated. In the embodiment where global source line (e.g., GSL1 of FIG. 4c) is accessed through a source access select transistor (not shown in FIG. 4c), the access select transistor is off during programming, resulting in the voltage on local source line 455 following the voltage on local bit line 454 during program and program inhibit. The same is true for the embodiment where the voltage on the local source line is provided by its parasitic capacitance C represented by capacitor 460 in FIG. 4c. In the embodiment of FIG. 4c, where there is a global source line but not a source access select transistor, the voltage applied to the global source line 413-1 of the addressed string should preferably track the voltage of the addressed global bit line 414-1 during program and program-inhibit.

[0093] Each of the incrementally higher voltage programming pulses is followed by a read cycle to determine if TFTs 416L and 416R have reached their respective target threshold voltage. If so, the drain, source and body voltages are raised to 10V (alternatively, these voltages are floated to close to 10V) to inhibit further programming, while word line WL31 continues to program other addressed TFTs on the same plane that have not yet attained their target threshold voltages. This sequence terminates when all addressed TFTs have been read-verified to be correctly programmed In the case of MLC, programming of one of the multiple threshold voltage states can be accelerated by setting each addressed global bit line to one of several predetermined voltages (e.g., 0V, 1.5V, 3.0V, or 4.5V, representing the four distinct states of the 2-bit data to be stored), and then applying the stepped programming pulses (up to around 20V) to word line WL31. In this manner, the addressed TFT receives a predetermined one of the effective tunneling voltages (i.e., 20, 18.5, 17, and 15.5 volts, respectively), resulting in one of predetermined threshold voltages being programmed into a TFT in a single programming sequence. Fine programming pulses may be subsequently provided at the individual TFT level.

Accelerated Whole-plane Parallel Programming

[0094] Because of the parasitic capacitance C intrinsic to every local source line in a multi-gate vertical NOR string array, all local source lines in a multi-gate vertical NOR string array can have 0V (for program) or 10V (for inhibit) momentarily placed (e.g., through global bit line GBL1 and bit line access string select transistor 411 and pre-charge transistor 470) on all vertical NOR strings in advance of applying the high voltage pulsing sequence. This procedure may be carried out by addressing the word line planes plane-by-plane. For each addressed word line plane, the programming pulsing sequence may be applied to many or all word lines on the addressed word line plane, while holding all word lines on the other word line planes at 0V, so as to program in parallel a large number of TFTs on the addressed plane, followed by individual read-verify, and where necessary, resetting the local source line of a properly programmed TFT into program-inhibit voltage. This approach provides a significant advantage, as programming time is relatively long (i.e., around 100 microsecond), while pre-charging all local source line capacitors or read-verifying all TFTs sharing the addressed word line plane is more than 1,000 times faster. Therefore, it pays to parallel program as many TFTs as possible in each word line plane. This accelerated programming feature provides even greater advantage in MLC programming which is considerably slower than single bit programming

Erase Operation

[0095] For some charge -trapping materials, the erase operation is performed by reverse-tunneling of the trapped charge, which can be rather slow, sometimes requiring tens of milliseconds of 20V or higher pulsing. Therefore, the erase operation may be implemented at the vertical NOR string array level (“block erase”), often performed in the background. A typical vertical NOR string array may have 64 word line planes, with each word line plane controlling, for example, 16,384×16,384 TFTs, for a total of approximately seventeen billion TFTs. A one-terabit chip may therefore include approximately 30 such vertical NOR string arrays, if two bits of data are stored on each TFT. In some embodiments, block erase may be carried out by applying around 20V to the P− channel shared by all TFTs in a vertical NOR string (e.g., body connection 456 in FIG. 4c and contact 556 in FIG. 5), while holding all word lines in the block at 0V. The duration of the erase pulse should be such that most TFTs in the block are erased to a slight enhancement mode threshold voltage, i.e., between zero and one volt. Some TFTs will overshoot and be erased into depletion mode (i.e., a slightly negative threshold voltage). A soft programming may be required to return the over-erased TFTs back into a slight enhancement mode threshold voltage after the termination of the erase pulses, as part of the erase command Vertical NOR strings that may include one of more depletion mode TFTs that cannot be programmed into enhancement mode may have to be retired, to be replaced by spare strings.

[0096] Alternatively, rather than providing the erase pulses to the body (i.e., the P− layer), the local source lines and the local bit lines (e.g., local source line 455 and local bit line 454 in FIG. 4c) on all vertical NOR string pairs in the vertical NOR string array are raised to around 20V, while holding all word lines on all word line planes at 0V for the duration of the erase pulse. This scheme requires that the global source line and the global bit line select decoders employ high voltage transistors that can withstand the 20V at their junctions. Alternatively, all TFTs sharing an addressed word line plane can be erased together by applying −20V pulses to all word lines on the addressed plane, while holding word lines on all other planes at 0V. All other voltages in the vertical NOR string pairs are held at 0V. This will erase only the X-Y slice of all TFTs touched by the one addressed plane of word lines.

Semi Non-Volatile NOR TFT Strings

[0097] Some charge-trapping materials (e.g., oxide-nitride-oxide or “ONO”) suitable for use in the vertical NOR string have long data retention time, typically in the order of many years, but relatively low endurance (i.e., performance degrades after some number of write-erase cycles, typically of the order of ten thousand cycles or less). However, in some embodiments one may select charge-trapping materials that store charge for much reduced retention times, but with much increased endurances (e.g., retention times in order of minutes or hours, endurance in the order of tens of millions of write-erase cycles). For example, in the embodiment of FIG. 7c, the tunnel dielectric layer 732c, typically a 6-8 nanometer layer of SiO₂, can be reduced in thickness to around 2 nanometers, or be replaced by another dielectric material (e.g., SiN) of similar thickness. The much thinner dielectric layer makes possible the use of modest voltages to introduce electrons by direct tunneling (as distinct from Fowler-Nordheim tunneling, which requires a higher voltage) into the charge-trapping layer, where they will be trapped from a few minutes to a few hours or days. Charge-trapping layer 732b can be silicon nitride, conductive nanodots dispersed in a thin dielectric film, or a combination of other charge-trapping films, including isolated thin floating gates. Blocking layer 732a can be silicon dioxide, aluminum oxide, hafnium oxide, silicon nitride, a high dielectric constant dielectric, or any combination thereof. Blocking layer 732a blocks electrons in charge-trapping layer 732b from escaping to the control gate word line. Trapped electrons will eventually leak out back into active region 730R, either as a result of the breakdown of the ultra-thin tunnel dielectric layer, or by reverse direct tunneling. However, such loss of trapped electrons is relatively slow. One may also use other combinations of charge storage materials, resulting in a high endurance but low retention “semi-volatile” storage TFT that requires periodic write or read refresh operations to replenish the lost charge. Because the vertical NOR strings of the present invention have a relatively fast read access (i.e. low latency), they may be used in some applications that currently require the use of dynamic random access memories (DRAMs). The vertical NOR strings of the present invention have significant advantages over DRAMs, having a much lower cost-per-bit, as DRAMs cannot be built in three dimensional stacks, and having a much lower power dissipation, as the refresh cycles need only be run approximately once every few minutes or every few hours, as compared to every few milliseconds required to refresh DRAMs. The three-dimensional semi-volatile storage TFTs of the present invention are achieved by selecting an appropriate material, such as those discussed above, for the charge-trapping material and by appropriately adapting the program/read/program-inhibit/erase conditions and incorporating the periodic data refreshes.

NROM/Mirror Bit NOR TFT Strings

[0098] In another embodiment of the current invention, the vertical NOR strings may be programmed using a channel hot-electron injection approach, similar to that which is used in two-dimensional NROM/Mirror Bit transistors, known to a person skilled in the art. Using the embodiment of FIG. 4a as an example, programming conditions for channel hot-electron injection may be: 8V on control gate 423p, 0V on local source line 455 and 5V on local drain line 454. Charge representing one bit is stored in the charge storage layer at one end of channel region 456 next to the junction with local bit line 454. By reversing polarity of local source line 455 and local bit line 454, charge representing a second bit is programmed and stored in the charge storage layer at the opposite end of channel region 456 next to the junction with local source line 455. Reading both bits requires reading in reverse order of the programming, as is well known to those skilled in the art. Channel hot-electron programming is much less efficient than programming by direct tunneling or Fowler-Nordheim tunneling and therefore it does not lend itself to the massively parallel programming possible with tunneling. However, each TFT has twice the bit density, making it attractive for applications such as archival memory. Erase for the NROM TFT embodiment can be achieved by employing the conventional NROM erase mechanism of band to band tunneling-induced hot-hole injection to neutralize the charge of the trapped electrons: apply −5V on the word line, 0V to local source line 455 and 5V to local bit line 454. Alternatively, the NROM TFT can be erased by applying a high positive substrate voltage V_bb to body region 456 with the word line at 0V. Because of the high programming current attendant to channel hot electron injection programming, all embodiments of vertical NROM TFT strings must employ hard-wired local source line and local bit line, such as in the embodiments of FIGS. 3a and 6c.

[0099] The above detailed description is provided to illustrate specific embodiments of the present invention and is not intended to be limiting. Numerous variations and modification within the scope of the present invention are possible. The present invention is set forth in the accompanying claims.